Shoppers say they would hand over the groceries first. The firms building the software expect them to start with financial products.

When a new technology disappoints, the problem is usually not the technology. It is that the people building it guessed wrong about what the user wanted, and nobody checked.

In June, Checkout.com published research on agentic commerce, the industry's term for software that finds, compares and buys on a person's behalf. It surveyed consumers across six markets and merchants in Britain and America. The results do not line up, and the way they fail to line up is specific enough to be useful.

Start with consent. One in four consumers, twenty four per cent, said they will never delegate a purchase to AI. Twenty seven per cent said they trust no organisation at all to run a shopping agent on their behalf. These are not people expressing caution about a particular brand. They are declining the category.

Then the money. Consumers said they would let an agent spend an average of one hundred and seventy seven pounds per purchase before wanting to approve it themselves. Merchants assumed the figure was two hundred. A twenty three pound gap sounds trivial until you notice which direction it runs in: the people building the permission system have set the dial looser than the people who have to live inside it.

The conditions consumers attached are worth reading as a group. Thirty per cent named spending caps as a non-negotiable, twenty nine per cent instant revocation, twenty eight per cent easy cancellation. Every one of those is a brake. Nobody asked for the agent to be cleverer. They asked to be able to stop it.

And then the part that does not fit at all. Asked what they would actually delegate, consumers chose groceries, at forty one per cent, and household supplies, at thirty one. Financial services came last, at fifteen. Checkout.com reports that this is at odds with merchants, who expect agentic commerce to take hold first in complex decisions such as financial products.

That is not a rounding difference. It is two groups holding opposite theories of what the machine is for. Merchants think its value is judgement, so it should handle the hard purchases. Consumers think its value is tedium, so it should handle the boring ones. Only one of those beliefs is being built into product roadmaps.

A necessary flag. Checkout.com sells payment infrastructure for exactly this market and has a commercial interest in it existing. Every figure above is theirs and is reported as theirs. The reason the research survives that conflict is that its headline finding, three per cent of transactions currently involving an agent against eighty nine per cent of merchants preparing, is not a number a vendor publishes to sell a boom.

WHY IT MATTERS HERE: The behaviour being designed for is delegation of judgement. The behaviour being offered is delegation of chores. If British retailers build for the first and their customers turn up wanting the second, the investment lands in the wrong place, and the loudest evidence that this is happening was published by a company that would rather it were not.